ProgrammableMarketing/dedaluslabs.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 7
  • This quarter · 10
  • Big bets · 3
  • Wildcards · 2
  • How we can help
Generated Nov 10, 2026 1 view
  1. Programmable
  2. Playbooks
  3. Dedalus Labs
Dedalus Labs logo

growth playbook · Nov 10, 2026

dedaluslabs.ai (opens in a new tab)

Dedalus Labs growth playbook

Diagnosis

Dedalus Labs (YC S25, $11M seed) has repositioned from an MCP gateway and agent SDK to persistent VMs for agents, and its engineering is strong: a custom hypervisor, a Rust VMM, sub-50ms starts and four open engineering roles. The marketing site has not caught up.

Its pages reach search engines and AI assistants as empty shells with a garbled title, pricing gives no unit rates, the speed claim is worded two ways, the old SDK audience (over 1,000 GitHub stars) is never told about Machines, and the startup page is unlisted with no offer.

Rivals E2B, Daytona and Blaxel publish rates, logos, compliance and $100-$200 credits. This playbook makes the site readable first, then adds clear pricing, one speed number, a migration path for the old audience and a real startup offer, then comparison pages, recipes, proof and engineering-led content, and finally templates, partner embeds and an open benchmark.

Growth Score

6/10

Biggest Opportunity

Developers and their coding agents now pick infrastructure by asking search and AI assistants; a site that serves real text, clear unit prices and one provable speed number can win those comparisons against better-funded rivals.

Biggest Risk

E2B, Daytona, Blaxel and Fly.io compete for the same agent workloads with bigger credits, published rates, compliance badges and named customers, and the team is hiring engineers only, so marketing work competes with product work for the founders' time.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Put the words back on dedaluslabs.ai: the homepage, pricing and blog reach search engines and AI assistants as empty pages

    Fetch the homepage with JavaScript turned off and list which components render only in the browser.

    High impact First result in 2-4 weeks
  2. 2

    Publish per-unit rates and show the $20 credit as machine-hours

    Write the three unit rates and the hours the $20 credit buys at the default 1 vCPU / 4 GiB size.

    High impact First result in 2 weeks
  3. 3

    Pick one start-time number and prove it with a public method

    Decide which interval the 50ms refers to and write the one-sentence definition.

    Medium impact First result in 2 weeks

02

Your idea portfolio

22 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 8 ideas are marked Unexpected: plays you are unlikely to have tried.

    1 of these are things Programmable can build and run for you.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 4Big bets · 8Fill-ins · 8Reconsider · 2
Effort →

Showing 22 of 22 ideas

This week 7

Quick wins with a clear first action.

  • 01

    When a crawler or AI assistant requests dedaluslabs.ai, it gets a title, a one-line meta description and schema.org data, and nothing else: no headline, no product copy, no links. The same is true of /pricing, /blog, /articles, /careers, /startup and individual posts such as /blog/sandbox-cold-start, and the /blog index links to none of its 15 posts. Search engines fill the gap badly: the homepage, blog, posts and careers page are listed as "DedalusDedalusDedalus Labs". In the same test E2B's homepage returned its full copy. Developers increasingly ask search and AI assistants which sandbox to use, and today those systems can only read Dedalus's metadata.

    High impactLow effort First result in 2-4 weeks Co Founder & CTO

    First action

    Fetch the homepage with JavaScript turned off and list which components render only in the browser.

    How to measure it

    Non-branded organic sessions

    Now
    –
    Target
    3x
    By
    90 days

    Steps

    1. 1

      Audit rendering

      Load each sitemap URL without JavaScript and record what text and links appear.

      Co-Founder & CTO half a day

    2. 2

      Render text on the server

      Move headline, copy, pricing, FAQ and post bodies to server or static rendering; keep effects client-side.

      Software Engineer, Infrastructure & Performance 3 days

    3. 3

      Fix the blog index

      Plain links to all 15 posts, with dates and summaries.

      Software Engineer, Infrastructure & Performance half a day

    4. 4

      Fix the wordmark

      One logo image with alt text "Dedalus Labs" instead of repeated text.

      Software Engineer, Infrastructure & Performance 1 hour

    5. 5

      Add a check

      A CI test that fails if key pages ship without their H1 and body text.

      Software Engineer, Infrastructure & Performance half a day

    6. 6

      Re-index

      Submit the sitemap in Search Console and request indexing of the top pages.

      Co-Founder & CEO 1 hour

    7. 7

      Check AI answers

      Ask the main AI assistants for agent sandbox options and record whether Dedalus is described correctly.

      Co-Founder & CEO monthly

    First result
    2-4 weeks
    Cost
    About a week of engineering time
    Owner
    Co Founder & CTO

    Smallest test

    Server-render /blog/ai-agent-sandbox-cost alone and watch its search impressions for two weeks.

    Why it works

    Render the marketing pages' text in the HTML the server sends, so it is there before any JavaScript runs: headline, product copy, pricing tables, FAQ, blog post bodies and the blog index with real links to every post. On Next.js this means server or static rendering for these routes, with the animated background and other effects loaded on top after the text. Make sure the wordmark is a single image with alt text so the title is not read as "Dedalus" three times. Then request re-indexing and check how the pages appear in search and in AI assistants.

    Search engines and AI assistants rank and quote what they can read. The blog already covers the exact questions buyers ask (sandbox cost, cold starts, VM versus isolate), but none of it is readable to them today, so the fix turns existing writing into traffic at almost no cost.

    Ready to use

    ChecklistReadable-site checklist

    [ ] Homepage H1 and product copy in server HTML
    [ ] /pricing tiers, rates and FAQ in server HTML
    [ ] /blog index links to every post
    [ ] Each post body in server HTML with date and author
    [ ] /startup, /careers, /articles copy in server HTML
    [ ] Wordmark as one image with alt text
    [ ] CI test for H1 + body on key pages
    [ ] Sitemap resubmitted, top 10 URLs re-indexed

    Risks and how to handle them

    • Server rendering slows the animated homepage

      Render text first and lazy-load the background; idea 7 measures the effect.

    • Search engines take weeks to re-crawl

      Request indexing for the ten most important URLs directly.

    Evidence

    • Your websiteHomepage, fetched twice on November 9, 2026: the HTML a crawler receives holds only the title "Fastest Persistent Computer for Agents | Dedalus Labs", the meta description "Persistent virtual machines for agents from Dedalus Labs." and schema.org data, with no headings, paragraphs or links (0 anchor links). /pricing, /blog, /articles, /careers, /startup, /store/vault and posts such as /blog/sandbox-cold-start return the same empty body, and the /blog index exposes no links to its 15 posts. In the same test, E2B's homepage returned its full copy.Source
    • Search resultsWeb search, November 9, 2026: the homepage, /blog, /blog/sandbox-cold-start, /blog/dedalus-seed-round and /careers are listed under the title "DedalusDedalusDedalus Labs" instead of their page titles; for the cold-start topic, the other results are rivals' guides (Blaxel, Unikraft).Source
    • Your websiteBlog: "AI Agent Sandbox Pricing in 2026" says "A 60-second session on a 2 vCPU / 4 GiB sandbox costs about a quarter of a cent in raw compute" and "The trap isn't the per-session price — it's paying for idle machines, over-provisioned CPU"; other posts cover sandbox cold starts, V8 isolates versus VMs and agent-infrastructure security posture.Source
  • 02

    Dedalus bills per second of active compute and prices storage separately, but no page, llms file or doc gives a rate for vCPU, memory or storage. Daytona publishes $0.0504 per vCPU-hour and $0.0162 per GiB-hour and offers $200 of free compute; E2B offers a $100 credit. Dedalus's $20 credit looks small next to those, even though its own blog says a 60-second 2 vCPU / 4 GiB session costs about a quarter of a cent, which puts $20 at roughly 130 hours of that machine.

    High impactLow effort First result in 2 weeks Co Founder & CEO

    First action

    Write the three unit rates and the hours the $20 credit buys at the default 1 vCPU / 4 GiB size.

    How to measure it

    Pricing-to-sign-up conversion

    Now
    –
    Target
    +25%
    By
    60 days

    Steps

    1. 1

      Agree the rates

      vCPU, memory and storage rates, and the free-sleep rule.

      Co-Founder & CEO 1 day

    2. 2

      Build the table

      Rates beside the Hobby, Pro and Enterprise tiers.

      Software Engineer, Infrastructure & Performance 1 day

    3. 3

      Build the calculator

      Size, active hours and machines in; monthly cost and credit hours out.

      Software Engineer, Infrastructure & Performance 2 days

    4. 4

      Add worked examples

      A coding agent per user, a nightly research, a CI runner.

      Co-Founder & CEO 1 day

    5. 5

      Sync the sources

      Same rates in llms-full.txt and the docs.

      Co-Founder & CTO 1 hour

    6. 6

      Measure

      Sign-ups from /pricing before and after.

      Co-Founder & CEO monthly

    First result
    2 weeks
    Cost
    Staff time only
    Owner
    Co Founder & CEO

    Smallest test

    Add one line under Hobby: "$20 runs a 2 vCPU / 4 GiB machine for about 130 active hours."

    Why it works

    Add a rate table (vCPU-second, GiB-second of memory, GiB-month of storage) and a small calculator to /pricing: machine size, active hours a day, number of machines, with the monthly cost and how long the $20 credit lasts. Show sleep at $0 compute next to it. Put the same rates in llms-full.txt and the docs.

    Developers compare unit prices before they sign up, and a credit expressed in hours of real work beats a smaller dollar figure. Clear rates also let AI assistants answer "how much does Dedalus cost" correctly.

    Ready to use

    CopyPricing table lines

    Compute: $__ per vCPU-hour, billed per second while running
    Memory: $__ per GiB-hour, billed per second while running
    Storage: $__ per GiB-month (first __ GiB included)
    Sleeping machines: $0 compute
    Hobby: $20 one-time credit = about __ hours of a 1 vCPU / 4 GiB machine

    Risks and how to handle them

    • Rates invite direct price comparisons

      Show total cost for a realistic agent day, where free sleep wins.

    Evidence

    • Your websitePricing page: meta description "Dedalus Machines pricing: $20 sign-up credit, Pro at $20/month, and per-second billing for active compute. Idle compute is free."; Hobby $0 with a "$20 one-time sign-up credit. No credit card required."; Pro "$20/month with $20 in monthly compute credits"; Enterprise "Custom pricing, dedicated support, SLAs, custom limits" (llms-full.txt). Storage is "priced separately", and neither the site, llms-full.txt nor the docs publish a per-unit rate for vCPU, memory or storage.Source
    • Competitor researchDaytona homepage: "Run AI Code."; "Sub 90ms sandbox creation"; "$200 in free compute included"; published rates of $0.0504 per vCPU-hour and $0.0162 per GiB of memory per hour; nine named customer logos and six testimonials; HIPAA, SOC 2 and GDPR claims.Source
    • Competitor researchE2B homepage: "The AI Agent Cloud"; Firecracker microVMs with their own kernel; pause and resume from snapshots and fork into up to 100 copies; SOC 2 Type II; US, EU and APAC regions; Hobby with one-time $100 in usage credits and up to 1 hour of continuous runtime per sandbox; Pro at $150 a month plus usage with 24 hours of runtime.Source
    • Your websiteBlog: "AI Agent Sandbox Pricing in 2026" says "A 60-second session on a 2 vCPU / 4 GiB sandbox costs about a quarter of a cent in raw compute" and "The trap isn't the per-session price — it's paying for idle machines, over-provisioned CPU"; other posts cover sandbox cold starts, V8 isolates versus VMs and agent-infrastructure security posture.Source
  • 03

    Speed is the headline claim, but it is worded two ways in the same file: "start in under 50ms" in the product section and "spins up in 50ms" in the company overview. Rivals each lead with a number: Daytona "Sub 90ms sandbox creation", Blaxel about 25ms resume. A buyer comparing them cannot tell whether Dedalus measures boot, wake or time to first command.

    Medium impactLow effort First result in 2 weeks Co Founder & CTO

    First action

    Decide which interval the 50ms refers to and write the one-sentence definition.

    How to measure it

    Benchmark script runs

    Now
    –
    Target
    200
    By
    60 days

    Steps

    1. 1

      Define it

      Which interval, which machine size, P50 and P99.

      Co-Founder & CTO 1 day

    2. 2

      Measure

      1,000 runs for fresh and woken machines.

      Systems Engineer 2 days

    3. 3

      Publish the script

      A public repo that reproduces the numbers with the CLI.

      Systems Engineer 1 day

    4. 4

      Update every page

      Homepage, pricing, llms files, docs, cold-start post.

      Co-Founder & CTO half a day

    5. 5

      Write it up

      A short post on how the hypervisor gets there.

      Co-Founder & CTO 2 days

    6. 6

      Re-run

      Every release, with results on the page.

      Software Engineer, Infrastructure & Performance per release

    First result
    2 weeks
    Cost
    Engineering time
    Owner
    Co Founder & CTO

    Smallest test

    Make the two wordings in llms-full.txt identical today.

    Why it works

    Define the number precisely (for example P50 and P99 from API call to first command, for a woken machine and a fresh one), publish the method and a small script anyone can run, and use exactly that wording on the homepage, pricing, llms files, docs and the cold-start post.

    A precise, reproducible number is more credible than a round one, and it gives AI assistants a single figure to quote when they compare providers.

    Risks and how to handle them

    • The measured P99 is less flattering

      Publish P50 and P99 together; honesty is the point.

    Evidence

    • Your websitellms.txt and llms-full.txt: "Dedalus Machines start in under 50ms" in the product section but "spins up in 50ms" in the company overview; full Linux with "systemd, package managers, and root access"; files "persist across sleep and wake"; a custom hypervisor built on Cloud Hypervisor; an FAQ that names E2B, Daytona and Fly.io without specifics; an agent skill installed with "npx skills add https://docs.dedaluslabs.ai --skill dedalus-machines"; an $11M seed and Y Combinator S25; no customers named.Source
    • Competitor researchDaytona homepage: "Run AI Code."; "Sub 90ms sandbox creation"; "$200 in free compute included"; published rates of $0.0504 per vCPU-hour and $0.0162 per GiB of memory per hour; nine named customer logos and six testimonials; HIPAA, SOC 2 and GDPR claims.Source
    • Competitor researchBlaxel homepage: "Infrastructure for autonomous agents"; about 25ms resume with memory state kept; up to $200 in credits with no card required; SOC 2, HIPAA and ISO 27001; 14 customer logos; the page says Blaxel has joined Baseten.Source
  • 04

    Dedalus's most-starred GitHub repos are from the earlier product: dedalus-agents-python (511 stars), dedalus-agents-typescript (284), wingman (180) and dedalus-mcp-python (138). The new Machines CLI and SDKs have 2 to 7 stars each. The dedalus-agents-python README still describes "the Dedalus REST API" and an MCP server with no mention of Machines. More than a thousand developers who already chose Dedalus once are never told what it sells now.

    Medium impactLow effort First result in 1 week Co Founder & CTO

    First action

    Open a README pull request on dedalus-agents-python with the banner and tagged link.

    How to measure it

    GitHub-referred sign-ups

    Now
    –
    Target
    100
    By
    60 days

    Steps

    1. 1

      Write the banner

      Two lines and a tagged quickstart link.

      Co-Founder & CEO 1 hour

    2. 2

      Decide support status

      Maintained, frozen or archived, stated plainly.

      Co-Founder & CTO 1 day

    3. 3

      Update READMEs

      The four legacy repos.

      Co-Founder & CTO half a day

    4. 4

      Pin new repos

      CLI and SDKs first on the organisation page.

      Co-Founder & CTO 10 minutes

    5. 5

      Update package pages

      Same note on the PyPI and npm descriptions.

      Co-Founder & CTO 1 hour

    6. 6

      Measure

      Clicks and sign-ups from the tagged links.

      Co-Founder & CEO monthly

    First result
    1 week
    Cost
    $0
    Owner
    Co Founder & CTO

    Smallest test

    Add the banner to dedalus-agents-python only and count clicks for two weeks.

    Why it works

    Add a short banner to the top of each legacy README ("Dedalus now builds persistent computers for agents" with a link to the quickstart and the $20 credit), state the support status of the old SDK plainly, and pin the Machines CLI and SDK repos on the organisation page. Track clicks with a tagged link.

    These developers build agents and already trust the brand; the cost of reaching them is one pull request per repo.

    Risks and how to handle them

    • Old SDK users feel abandoned

      Give a clear support date and a migration note.

    Evidence

    • GitHubGitHub organisation (84 public repos, 131 followers): the most-starred repos come from the earlier agent SDK: dedalus-agents-python (511 stars), dedalus-agents-typescript (284), wingman (180) and dedalus-mcp-python (138). The Machines tools are new and small: dedalus-cli (7), dedalus-typescript (5), dedalus-python (3) and dedalus-go (2), all updated November 6, 2026. The dedalus-agents-python README offers "convenient access to the Dedalus REST API" and links an MCP server, with no mention of Dedalus Machines.Source
    • Your websitePositioning: in November 2025 the homepage headline was "Connect any model to any MCP server with a single API."; today it is persistent VMs for agents. The seed-round post ($11M, "led by Kindred Ventures and Saga Ventures") gives the goal as helping "developers build complex agents across any model and tool in 5 lines of code", and the Dedalus Auth post describes "A multi-tenant auth layer that ensures MCP agents never handle raw secrets."Source
  • 05

    The homepage now sells persistent VMs, but the blog still describes itself as "about MCP, AI agents, and developer tooling", three May 2026 posts demo the earlier agent product, /store/vault is "An interactive preview from the Dedalus Archive", and /startup promises "marketplace distribution". Six posts, including the seed round, were re-dated to the same day. A visitor or AI assistant reading the site sees two companies.

    Medium impactLow effort First result in 1 month Co Founder & CEO

    First action

    List every page and mark it keep, reframe or redirect.

    How to measure it

    On-message pages

    Now
    –
    Target
    100%
    By
    30 days

    Steps

    1. 1

      Inventory

      All 24 sitemap URLs plus /startup.

      Co-Founder & CEO 2 hours

    2. 2

      Rewrite descriptions

      Blog, articles and startup meta text.

      Co-Founder & CEO 2 hours

    3. 3

      Annotate old posts

      A note at the top linking to Machines.

      Co-Founder & CEO 2 hours

    4. 4

      Fix dates

      Original publish dates, shown on each post.

      Software Engineer, Infrastructure & Performance 2 hours

    5. 5

      Decide on /store

      Explain it or redirect it.

      Co-Founder & CEO 1 day

    6. 6

      Recheck summaries

      How search and AI assistants describe Dedalus after re-crawl.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    $0
    Owner
    Co Founder & CEO

    Smallest test

    Change the blog description today.

    Why it works

    Rewrite the blog description around agent compute; add a one-line note to each MCP-era post that points to Machines; decide whether /store and /store/vault stay (and say what they are) or redirect; restore original publish dates on re-dated posts and show them on the page; drop "marketplace" from /startup unless it is still offered.

    One clear story converts better and stops search and AI summaries from describing Dedalus as an MCP gateway.

    Risks and how to handle them

    • Losing traffic from old posts

      Keep and annotate pages that bring visitors; redirect only empty ones.

    Evidence

    • Your websiteSitemap (24 URLs, November 9, 2026): 15 blog posts, six re-dated November 4, 2026 (seed round, Dedalus Auth launch, Machines update, "From Today to A2a", "What is an Agent", a YC interview); May 2026 posts from the earlier agent product ("Agent CRM Research", "Finance research", "One Click PDF Generator"); /store and /store/vault ("An interactive preview from the Dedalus Archive"); /private-beta. No comparison, use-case, customer-story or changelog pages. The blog description still reads "Technical writing, tutorials, and company updates about MCP, AI agents, and developer tooling." A /startup page ("Credits, support, deployment tooling, and marketplace distribution for startups building production AI agents on Dedalus") is live but missing from the sitemap and states no offer.Source
    • Your websitePositioning: in November 2025 the homepage headline was "Connect any model to any MCP server with a single API."; today it is persistent VMs for agents. The seed-round post ($11M, "led by Kindred Ventures and Saga Ventures") gives the goal as helping "developers build complex agents across any model and tool in 5 lines of code", and the Dedalus Auth post describes "A multi-tenant auth layer that ensures MCP agents never handle raw secrets."Source
  • 06

    A "Dedalus for Startups" page already exists, promising "Credits, support, deployment tooling", but it is not in the sitemap, its body reaches crawlers empty, and it states no amount, eligibility or way to apply. Meanwhile E2B gives $100 of credit, and Daytona and Blaxel each offer up to $200. Startups building agents compare these offers first, and Dedalus's public number is a $20 credit.

    High impactLow effort First result in 3 weeks Co Founder & CEO

    First action

    Decide the credit amount and eligibility in one paragraph.

    How to measure it

    Startups onboarded

    Now
    –
    Target
    40
    By
    90 days

    Steps

    1. 1

      Set the offer

      Credit, perks, eligibility and duration.

      Co-Founder & CEO 1 day

    2. 2

      Write the page

      Offer, who qualifies, how to apply, what happens next.

      Co-Founder & CEO 1 day

    3. 3

      List it

      Sitemap, footer, pricing page and docs.

      Software Engineer, Infrastructure & Performance 1 hour

    4. 4

      Automate credits

      Approved applicants get credit without manual work.

      Co-Founder & CTO 2 days

    5. 5

      Distribute

      YC deals board and the seed investors' portfolio channels.

      Co-Founder & CEO 1 week

    6. 6

      Track

      Applications, active machines and paid conversion.

      Co-Founder & CEO monthly

    First result
    3 weeks
    Cost
    Compute credits
    Owner
    Co Founder & CEO

    Smallest test

    Post the offer to the YC deals board only.

    Why it works

    Set a concrete startup offer (for example a credit sized in machine-hours, Pro features and a shared Slack or Discord channel with the engineers) with simple eligibility: pre-Series A, building agents. Publish it on /startup with an application form, add it to the sitemap and footer, and offer it through YC and the seed investors' portfolios.

    Startups that build on a provider early stay as they grow, and a visible offer turns the existing page into a pipeline at small cost, because sleeping machines cost nothing.

    Risks and how to handle them

    • Credit abuse

      Require a company domain and cap credits per month.

    Evidence

    • Your websiteSitemap (24 URLs, November 9, 2026): 15 blog posts, six re-dated November 4, 2026 (seed round, Dedalus Auth launch, Machines update, "From Today to A2a", "What is an Agent", a YC interview); May 2026 posts from the earlier agent product ("Agent CRM Research", "Finance research", "One Click PDF Generator"); /store and /store/vault ("An interactive preview from the Dedalus Archive"); /private-beta. No comparison, use-case, customer-story or changelog pages. The blog description still reads "Technical writing, tutorials, and company updates about MCP, AI agents, and developer tooling." A /startup page ("Credits, support, deployment tooling, and marketplace distribution for startups building production AI agents on Dedalus") is live but missing from the sitemap and states no offer.Source
    • Competitor researchE2B homepage: "The AI Agent Cloud"; Firecracker microVMs with their own kernel; pause and resume from snapshots and fork into up to 100 copies; SOC 2 Type II; US, EU and APAC regions; Hobby with one-time $100 in usage credits and up to 1 hour of continuous runtime per sandbox; Pro at $150 a month plus usage with 24 hours of runtime.Source
    • Competitor researchDaytona homepage: "Run AI Code."; "Sub 90ms sandbox creation"; "$200 in free compute included"; published rates of $0.0504 per vCPU-hour and $0.0162 per GiB of memory per hour; nine named customer logos and six testimonials; HIPAA, SOC 2 and GDPR claims.Source
    • Competitor researchBlaxel homepage: "Infrastructure for autonomous agents"; about 25ms resume with memory state kept; up to $200 in credits with no card required; SOC 2, HIPAA and ISO 27001; 14 customer logos; the page says Blaxel has joined Baseten.Source
  • 07

    A company selling speed has a slow homepage: real Chrome visitors fail Core Web Vitals on LCP (p75 4.5s, with 2.2s to first byte). The page weighs 5.1 MB, with 2.3 MB of JavaScript, about 1 MB of it unused, and its largest element is a fixed background canvas. Tag manager, Facebook and Reddit tags add about 770 KB.

    Medium impactMedium effort First result in 4 weeks Software Engineer

    First action

    Run a bundle analysis on the homepage route.

    How to measure it

    Real-visitor LCP (p75)

    Now
    –
    Target
    under 2.5s
    By
    60 days

    Steps

    1. 1

      Analyse the bundle

      Find the largest unused modules.

      Software Engineer, Infrastructure & Performance half a day

    2. 2

      Defer the canvas

      Static first frame; start the animation after load.

      Software Engineer, Infrastructure & Performance 1 day

    3. 3

      Split the code

      Route-level chunks; drop legacy polyfills.

      Software Engineer, Infrastructure & Performance 2 days

    4. 4

      Review tags

      Owner and purpose for each marketing tag; delay or remove.

      Co-Founder & CEO 1 hour

    5. 5

      Fix caching and first byte

      Long cache lifetimes for static files; check why first byte takes 2.2s.

      Software Engineer, Infrastructure & Performance 1 day

    6. 6

      Set a budget

      CI fails if homepage JavaScript grows past the limit.

      Software Engineer, Infrastructure & Performance half a day

    First result
    4 weeks (field data lags)
    Cost
    Engineering time
    Owner
    Software Engineer, Infrastructure & Performance

    Smallest test

    Replace the canvas with a static image on mobile only and compare lab LCP.

    Why it works

    Serve the hero text first (idea 1 helps), replace the canvas background with a static image or CSS until the page is idle, split the JavaScript so the homepage loads only what it uses, cache static assets, and review the marketing tags: keep the ones someone uses, load them after interaction, and remove duplicates.

    Developers notice slow pages from a performance company, and a passing LCP also helps search ranking.

    Risks and how to handle them

    • Losing the visual identity

      Keep the animation, just start it after the content.

    Builds on

    Evidence

    • Google PageSpeedHomepage, November 9, 2026: real Chrome visitors fail Core Web Vitals on LCP (p75 4.5s, with time to first byte at 2.2s); mobile lab score 37 (desktop 61) with a lab LCP of 16.5s; 5.1 MB over 97 requests, including 2.3 MB of JavaScript of which about 1,018 KB is unused (about 3.9s of savings); the largest element is a fixed background canvas; tag manager (461 KB), Facebook (282 KB) and Reddit (27 KB) tags add about 770 KB.Source
    • Your websiteHomepage, fetched twice on November 9, 2026: the HTML a crawler receives holds only the title "Fastest Persistent Computer for Agents | Dedalus Labs", the meta description "Persistent virtual machines for agents from Dedalus Labs." and schema.org data, with no headings, paragraphs or links (0 anchor links). /pricing, /blog, /articles, /careers, /startup, /store/vault and posts such as /blog/sandbox-cold-start return the same empty body, and the /blog index exposes no links to its 15 posts. In the same test, E2B's homepage returned its full copy.Source

This quarter 10

The core plays for the next three months.

  • 08

    Dedalus's own FAQ already names E2B, Daytona and Fly.io as the alternatives buyers ask about, but answers without specifics, and the sitemap has no comparison pages. The real differences are concrete: E2B caps Hobby sandboxes at 1 hour and Pro at 24 hours, Daytona creates sandboxes in under 90ms, Dedalus machines persist with free sleep and no stated runtime cap.

    High impactMedium effort First result in 1-2 months Co Founder & CEO

    First action

    Draft the E2B table from both public sites.

    How to measure it

    Comparison-page sign-ups

    Now
    –
    Target
    50 a month
    By
    90 days

    Steps

    1. 1

      Pick rivals

      E2B, Daytona, Fly.io; then Blaxel.

      Co-Founder & CEO 1 hour

    2. 2

      Check facts

      Rivals' current public pricing and limits, with links.

      Co-Founder & CEO 2 days

    3. 3

      Write migration snippets

      Equivalent create, exec and sleep calls.

      Systems Engineer 2 days

    4. 4

      Write pages

      Tables, honest trade-offs, date stamp.

      Co-Founder & CEO 1 week

    5. 5

      Review claims

      Trademark use and factual accuracy.

      Co-Founder & CEO 2 days

    6. 6

      Refresh

      Re-check rival pages every quarter.

      Co-Founder & CEO quarterly

    First result
    1-2 months
    Cost
    Staff time only
    Owner
    Co Founder & CEO

    Smallest test

    Publish the E2B page alone and track its sign-ups.

    Why it works

    Write one page per rival with a dated, sourced table: isolation model, start and resume time, persistence, runtime limits, sleep billing, unit prices, credits, SDKs, compliance and regions. Be honest where the rival wins (regions, compliance today) and end with a migration snippet showing the equivalent Dedalus CLI or SDK calls.

    Comparison searches come from developers about to choose, and fair tables are what AI assistants quote when asked "E2B vs Dedalus".

    Risks and how to handle them

    • Rival pricing changes

      Date every table and link the source.

    Builds on

    Evidence

    • Your websitellms.txt and llms-full.txt: "Dedalus Machines start in under 50ms" in the product section but "spins up in 50ms" in the company overview; full Linux with "systemd, package managers, and root access"; files "persist across sleep and wake"; a custom hypervisor built on Cloud Hypervisor; an FAQ that names E2B, Daytona and Fly.io without specifics; an agent skill installed with "npx skills add https://docs.dedaluslabs.ai --skill dedalus-machines"; an $11M seed and Y Combinator S25; no customers named.Source
    • Competitor researchE2B homepage: "The AI Agent Cloud"; Firecracker microVMs with their own kernel; pause and resume from snapshots and fork into up to 100 copies; SOC 2 Type II; US, EU and APAC regions; Hobby with one-time $100 in usage credits and up to 1 hour of continuous runtime per sandbox; Pro at $150 a month plus usage with 24 hours of runtime.Source
    • Competitor researchDaytona homepage: "Run AI Code."; "Sub 90ms sandbox creation"; "$200 in free compute included"; published rates of $0.0504 per vCPU-hour and $0.0162 per GiB of memory per hour; nine named customer logos and six testimonials; HIPAA, SOC 2 and GDPR claims.Source
    • Your websiteSitemap (24 URLs, November 9, 2026): 15 blog posts, six re-dated November 4, 2026 (seed round, Dedalus Auth launch, Machines update, "From Today to A2a", "What is an Agent", a YC interview); May 2026 posts from the earlier agent product ("Agent CRM Research", "Finance research", "One Click PDF Generator"); /store and /store/vault ("An interactive preview from the Dedalus Archive"); /private-beta. No comparison, use-case, customer-story or changelog pages. The blog description still reads "Technical writing, tutorials, and company updates about MCP, AI agents, and developer tooling." A /startup page ("Credits, support, deployment tooling, and marketplace distribution for startups building production AI agents on Dedalus") is live but missing from the sitemap and states no offer.Source
  • 09

    The quickstart is four steps (install the CLI, create an API key in the dashboard, create a machine, run a command), but nothing on the site or docs says how long it takes. Every new account gets $20 of credit, so the moment that matters is the first command that runs.

    High impactMedium effort First result in 1 month Co Founder & CTO

    First action

    Add events for the five funnel steps.

    How to measure it

    10-minute activation

    Now
    –
    Target
    60%
    By
    90 days

    Steps

    1. 1

      Instrument

      Five funnel events with timestamps.

      Software Engineer, Infrastructure & Performance 2 days

    2. 2

      Find the drop

      Where most new users stop.

      Co-Founder & CTO 1 week

    3. 3

      Prefill the command

      Copy-ready create and exec with the key in place.

      Software Engineer, Infrastructure & Performance 2 days

    4. 4

      Browser terminal

      Open a shell on the first machine from the dashboard.

      Systems Engineer 2 weeks

    5. 5

      Nudge

      Email or CLI hint when a key has no machine after an hour.

      Co-Founder & CEO 2 days

    6. 6

      Review weekly

      Activation by sign-up source.

      Co-Founder & CEO weekly

    First result
    1 month
    Cost
    Engineering time
    Owner
    Co Founder & CTO

    Smallest test

    Show the prefilled command after sign-up for half of new users.

    Why it works

    Instrument the funnel (sign-up, key created, first machine, first command, machine used on day 7), then cut steps: show a ready-to-copy command with the key filled in right after sign-up, offer a browser terminal on a first machine, and send a short email or CLI hint to anyone who created a key but no machine within an hour.

    Developer products win or lose in the first ten minutes, and the team can see exactly where new users stop.

    Risks and how to handle them

    • Browser terminal raises abuse risk

      Limit it to the first machine with the default size.

    Evidence

    • DocsMachines are Linux VMs with their own kernel and root access that sleep "after five minutes of inactivity by default"; a CLI, an HTTP API and Python, TypeScript and Go SDKs; a four-step quickstart (install the CLI, set an API key, create a machine, run a command) with a default of 1 vCPU, 4,096 MiB and 10 GiB; "you only pay for the compute you use while the machine is running", with no unit rates. The docs index lists no cookbook, examples, pricing or limits pages, and a cookbook recipe still listed in search (agent-sandbox-per-user) returns 404.Source
    • Your websitePricing page: meta description "Dedalus Machines pricing: $20 sign-up credit, Pro at $20/month, and per-second billing for active compute. Idle compute is free."; Hobby $0 with a "$20 one-time sign-up credit. No credit card required."; Pro "$20/month with $20 in monthly compute credits"; Enterprise "Custom pricing, dedicated support, SLAs, custom limits" (llms-full.txt). Storage is "priced separately", and neither the site, llms-full.txt nor the docs publish a per-unit rate for vCPU, memory or storage.Source
  • 10

    The docs cover the machine lifecycle well but list no cookbook, examples, pricing or limits pages, and a recipe search engines still show ("agent-sandbox-per-user") now returns 404. Developers search for patterns ("one sandbox per user", "run a coding agent safely"), not for lifecycle commands.

    High impactMedium effort First result in 1 month Systems Engineer

    First action

    Restore the per-user recipe so the 404 address works again.

    How to measure it

    Recipe-assisted sign-ups

    Now
    –
    Target
    25% of sign-ups
    By
    1 quarter

    Steps

    1. 1

      Restore the recipe

      Per-user sandbox at its old URL.

      Systems Engineer 1 day

    2. 2

      Pick five more

      From support questions and Discord.

      Co-Founder & CTO 1 day

    3. 3

      Build repos

      Each runs with one command.

      Systems Engineer Intern 3 weeks

    4. 4

      Write pages

      Why, code, cost estimate using the published rates.

      Systems Engineer 1 week

    5. 5

      Link them

      Homepage, docs index, comparison pages, READMEs.

      Co-Founder & CEO 1 day

    6. 6

      Test in CI

      Recipes run nightly so they never break silently.

      Infrastructure Engineer Intern 2 days

    First result
    1 month
    Cost
    Engineering time
    Owner
    Systems Engineer

    Smallest test

    Restore the 404 recipe and watch its traffic.

    Why it works

    Restore the per-user recipe at its old address and add five more, each a runnable repo with a docs page: a coding agent per user, a long-running research that sleeps between steps, a browser agent, a CI runner, a data notebook per analyst, and a multi-agent setup sharing files. Link them from the homepage, the comparison pages and the READMEs.

    Recipes rank for the task searches buyers make and cut the time from interest to a working agent.

    Risks and how to handle them

    • Recipes go stale

      Nightly CI runs each one.

    Evidence

    • DocsMachines are Linux VMs with their own kernel and root access that sleep "after five minutes of inactivity by default"; a CLI, an HTTP API and Python, TypeScript and Go SDKs; a four-step quickstart (install the CLI, set an API key, create a machine, run a command) with a default of 1 vCPU, 4,096 MiB and 10 GiB; "you only pay for the compute you use while the machine is running", with no unit rates. The docs index lists no cookbook, examples, pricing or limits pages, and a cookbook recipe still listed in search (agent-sandbox-per-user) returns 404.Source
    • HiringFour open roles on Ashby, all engineering in San Francisco: Software Engineer, Infrastructure & Performance (posted September 29, 2026: the work "can bring product releases and customer commitments forward by weeks"; an "in-office homelab"; internal tools Hollywood, which "generates GitHub Actions from typed TypeScript definitions", and the Bessemer build system), Systems Engineer (posted June 29, 2026: the control plane and the Rust VMM, "the core technology our customers depend on"), a Systems Engineer Intern and an Infrastructure Engineer Intern. No marketing, sales or developer-relations roles are open.Source
  • 11

    Dedalus already ships an agent skill ("npx skills add https://docs.dedaluslabs.ai --skill dedalus-machines"). Developers increasingly let a coding agent pick and set up tools, so the skill is a distribution channel, but it is mentioned only inside llms.txt, which crawlers and humans rarely open.

    High impactLow effort First result in 1 month Co Founder & CTO

    First action

    Add a tracked identifier to machines created through the skill.

    How to measure it

    Skill-created machines

    Now
    –
    Target
    30% of new machines
    By
    1 quarter

    Steps

    1. 1

      Tag skill traffic

      A client identifier on CLI calls made via the skill.

      Systems Engineer 1 day

    2. 2

      Give it a page

      What it does, one-line install, safety notes.

      Co-Founder & CTO 1 day

    3. 3

      Record a demo

      An agent spins up a machine and runs tests in it.

      Co-Founder & CEO 2 days

    4. 4

      List it

      Public skill directories and agent tool registries.

      Co-Founder & CEO 1 week

    5. 5

      Improve the instructions

      Fix the steps agents fail on, from logs.

      Systems Engineer ongoing

    6. 6

      Measure

      Installs, machines and paid conversion from the skill.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Co Founder & CTO

    Smallest test

    Put the install line on the homepage hero for two weeks.

    Why it works

    Give the skill its own page and a homepage line ("Tell your agent: install the Dedalus skill"), submit it to the public skill and MCP directories, add a short demo of an agent creating its own machine and running its tests there, and track installs and the machines created through it.

    When the agent is the one choosing infrastructure, the provider whose instructions it already has wins the default.

    Risks and how to handle them

    • Agents create machines users forget

      Keep the five-minute autosleep default and show it in the skill.

    Evidence

    • Your websitellms.txt and llms-full.txt: "Dedalus Machines start in under 50ms" in the product section but "spins up in 50ms" in the company overview; full Linux with "systemd, package managers, and root access"; files "persist across sleep and wake"; a custom hypervisor built on Cloud Hypervisor; an FAQ that names E2B, Daytona and Fly.io without specifics; an agent skill installed with "npx skills add https://docs.dedaluslabs.ai --skill dedalus-machines"; an $11M seed and Y Combinator S25; no customers named.Source
    • DocsMachines are Linux VMs with their own kernel and root access that sleep "after five minutes of inactivity by default"; a CLI, an HTTP API and Python, TypeScript and Go SDKs; a four-step quickstart (install the CLI, set an API key, create a machine, run a command) with a default of 1 vCPU, 4,096 MiB and 10 GiB; "you only pay for the compute you use while the machine is running", with no unit rates. The docs index lists no cookbook, examples, pricing or limits pages, and a cookbook recipe still listed in search (agent-sandbox-per-user) returns 404.Source
  • 12

    Dedalus sells isolation for untrusted agent code and has a Vanta trust center, but the public trust page lists no frameworks and nothing on the site states a compliance status. Rivals lead with it: E2B shows SOC 2 Type II, Blaxel SOC 2, HIPAA and ISO 27001, Daytona HIPAA, SOC 2 and GDPR. The blog already has a post on agent-infrastructure security posture.

    High impactMedium effort First result in 1-2 months Co Founder & CEO

    First action

    Write the one-line compliance status and publish it on the trust center.

    How to measure it

    Security reviews passed

    Now
    –
    Target
    80% of reviews
    By
    2 quarters

    Steps

    1. 1

      State status

      Frameworks achieved or in progress, with dates.

      Co-Founder & CEO 1 day

    2. 2

      Fill the trust center

      Controls, policies, subprocessors.

      Co-Founder & CTO 1 week

    3. 3

      Write the isolation page

      Own kernel, memory isolation, deletion, network rules.

      Systems Engineer 1 week

    4. 4

      Add to pricing

      Security row and trust link on Enterprise.

      Software Engineer, Infrastructure & Performance 1 day

    5. 5

      Answer the questionnaire once

      A standard security questionnaire on file.

      Co-Founder & CTO 1 week

    6. 6

      Track

      Deals that asked for security material and outcome.

      Co-Founder & CEO monthly

    First result
    1-2 months
    Cost
    Audit fees if not already under way
    Owner
    Co Founder & CEO

    Smallest test

    Add one compliance line and the trust link to the Enterprise tier.

    Why it works

    State the current compliance status plainly (for example "SOC 2 Type I in progress, report expected in Q_"), list the controls on the trust center, and add a security section to pricing and the Enterprise tier: own kernel per machine, VM-level isolation, the data-deletion policy, and the security post. Offer the trust center link in every Enterprise conversation.

    Teams moving agents to production need a security answer before a technical one; showing it early stops deals stalling at review.

    Risks and how to handle them

    • Claiming a status not yet achieved

      Say "in progress" with the auditor's timeline only.

    Evidence

    • Your websiteSupport and trust: a Vanta trust center at trust.dedaluslabs.ai whose public page lists no frameworks; an incident.io status page (operational); a Discord community; support by email only, with no chat or meeting booking; no case studies, testimonials or changelog on the site.Source
    • Competitor researchE2B homepage: "The AI Agent Cloud"; Firecracker microVMs with their own kernel; pause and resume from snapshots and fork into up to 100 copies; SOC 2 Type II; US, EU and APAC regions; Hobby with one-time $100 in usage credits and up to 1 hour of continuous runtime per sandbox; Pro at $150 a month plus usage with 24 hours of runtime.Source
    • Competitor researchBlaxel homepage: "Infrastructure for autonomous agents"; about 25ms resume with memory state kept; up to $200 in credits with no card required; SOC 2, HIPAA and ISO 27001; 14 customer logos; the page says Blaxel has joined Baseten.Source
    • Your websiteBlog: "AI Agent Sandbox Pricing in 2026" says "A 60-second session on a 2 vCPU / 4 GiB sandbox costs about a quarter of a cent in raw compute" and "The trap isn't the per-session price — it's paying for idle machines, over-provisioned CPU"; other posts cover sandbox cold starts, V8 isolates versus VMs and agent-infrastructure security posture.Source
  • 13

    Neither the site nor the llms files name a single customer, and there are no testimonials or case studies. Daytona shows nine customer logos and six testimonials; Blaxel shows 14 logos. The hiring posts say engineers build "the core technology our customers depend on", so the customers exist; they are just not visible.

    High impactMedium effort First result in 2 months Co Founder & CEO

    First action

    Rank accounts by active machine-hours and email the top five.

    How to measure it

    Customer stories live

    Now
    –
    Target
    3
    By
    90 days

    Steps

    1. 1

      Rank accounts

      By usage and fit.

      Co-Founder & CTO 1 day

    2. 2

      Ask

      Founder-to-founder request with a credit offer.

      Co-Founder & CEO 1 week

    3. 3

      Interview

      30 minutes each, with usage numbers.

      Co-Founder & CEO 2 weeks

    4. 4

      Write

      Problem, setup, numbers, quote.

      Co-Founder & CEO 2 weeks

    5. 5

      Approve

      Customer sign-off on text and logo.

      Co-Founder & CEO 2 weeks

    6. 6

      Place

      Homepage, pricing, comparison pages, llms files.

      Software Engineer, Infrastructure & Performance 1 day

    First result
    2 months
    Cost
    Compute credits as thanks
    Owner
    Co Founder & CEO

    Smallest test

    Get one quote approved and put it on the pricing page.

    Why it works

    Ask the three most active accounts for a short story each: the agent they run, why persistent machines, numbers (machines per day, cost versus before, start time). Offer credit for their time. Publish them as pages, use their logos on the homepage and pricing, and quote them on the comparison pages.

    For a new infrastructure provider, proof that someone like me runs production on it is the biggest objection to clear.

    Risks and how to handle them

    • Customers won't be named

      Publish anonymised stories with real numbers first.

    Evidence

    • Your websiteSupport and trust: a Vanta trust center at trust.dedaluslabs.ai whose public page lists no frameworks; an incident.io status page (operational); a Discord community; support by email only, with no chat or meeting booking; no case studies, testimonials or changelog on the site.Source
    • Competitor researchDaytona homepage: "Run AI Code."; "Sub 90ms sandbox creation"; "$200 in free compute included"; published rates of $0.0504 per vCPU-hour and $0.0162 per GiB of memory per hour; nine named customer logos and six testimonials; HIPAA, SOC 2 and GDPR claims.Source
    • HiringFour open roles on Ashby, all engineering in San Francisco: Software Engineer, Infrastructure & Performance (posted September 29, 2026: the work "can bring product releases and customer commitments forward by weeks"; an "in-office homelab"; internal tools Hollywood, which "generates GitHub Actions from typed TypeScript definitions", and the Bessemer build system), Systems Engineer (posted June 29, 2026: the control plane and the Rust VMM, "the core technology our customers depend on"), a Systems Engineer Intern and an Infrastructure Engineer Intern. No marketing, sales or developer-relations roles are open.Source
  • 14

    The Enterprise tier promises "Custom pricing, dedicated support, SLAs, custom limits", but the only contact on the site is [email protected], with no form, chat or meeting booking. A team ready to run thousands of agent machines has to write to the same inbox as a hobby user with a billing question.

    Medium impactLow effort First result in 1 month Co Founder & CEO

    First action

    Set up the booking link and an enterprise@ alias.

    How to measure it

    Enterprise calls booked

    Now
    –
    Target
    8 a month
    By
    90 days

    Steps

    1. 1

      Write the form

      Five qualifying questions.

      Co-Founder & CEO 1 hour

    2. 2

      Add booking

      A founder's calendar after submit.

      Co-Founder & CEO 1 hour

    3. 3

      Place the link

      Pricing, trust center, comparison pages, docs.

      Software Engineer, Infrastructure & Performance half a day

    4. 4

      Route separately

      Away from the support inbox.

      Co-Founder & CTO 1 hour

    5. 5

      Prepare the call

      Pricing sheet, security pack, limits.

      Co-Founder & CEO 1 day

    6. 6

      Track

      Calls, pilots and contracts.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    $0
    Owner
    Co Founder & CEO

    Smallest test

    Replace the Enterprise button's email link with a booking link.

    Why it works

    Add a short Enterprise form (expected concurrent machines, workload, region, security needs) that books a 30-minute call with a founder or engineer, route it separately from support, and reply within one business day. Link it from pricing, the trust center, the comparison pages and the docs' limits section.

    Big accounts pay for a startup's next stage, and a clear path with qualifying questions makes each call productive.

    Risks and how to handle them

    • Founder time

      Qualify on expected machine volume before booking.

    Evidence

    • Your websitePricing page: meta description "Dedalus Machines pricing: $20 sign-up credit, Pro at $20/month, and per-second billing for active compute. Idle compute is free."; Hobby $0 with a "$20 one-time sign-up credit. No credit card required."; Pro "$20/month with $20 in monthly compute credits"; Enterprise "Custom pricing, dedicated support, SLAs, custom limits" (llms-full.txt). Storage is "priced separately", and neither the site, llms-full.txt nor the docs publish a per-unit rate for vCPU, memory or storage.Source
    • Your websiteSupport and trust: a Vanta trust center at trust.dedaluslabs.ai whose public page lists no frameworks; an incident.io status page (operational); a Discord community; support by email only, with no chat or meeting booking; no case studies, testimonials or changelog on the site.Source
  • 15

    The job posts describe things infrastructure engineers love to read about: a custom Rust VMM and control plane, an in-office homelab of real machines and networks, Hollywood (which "generates GitHub Actions from typed TypeScript definitions", 18 stars) and the Bessemer build system. The buyers of agent infrastructure are the same engineers, and the team has no marketing or developer-relations hire.

    Medium impactMedium effort First result in 1 month Co Founder & CTO

    First action

    Pick the first topic and assign an author.

    How to measure it

    Deep-dive sign-ups

    Now
    –
    Target
    150 a quarter
    By
    2 quarters

    Steps

    1. 1

      Plan topics

      Six months of topics, one author each.

      Co-Founder & CTO 1 day

    2. 2

      Write

      Diagrams, numbers, code.

      Systems Engineer 1 week each

    3. 3

      Edit

      Clear opening, honest limits.

      Co-Founder & CEO 2 days each

    4. 4

      Publish

      Server-rendered with date and author.

      Software Engineer, Infrastructure & Performance 1 hour

    5. 5

      Distribute

      Hacker News, Reddit, X, Discord.

      Co-Founder & CEO 1 day

    6. 6

      Measure

      Visits, sign-ups and job applications per post.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    Engineering time
    Owner
    Co Founder & CTO

    Smallest test

    Publish the sub-50ms start write-up from idea 3.

    Why it works

    Publish one deep dive a month written by the engineer who built it: how a machine starts in under 50ms, how storage is split from compute, what the homelab catches that cloud tests miss, why Hollywood exists. Post to Hacker News, the relevant subreddits and X, and end each with a recipe and the $20 credit.

    Deep technical writing earns links and trust in a way ads cannot, and it doubles as recruiting for the four open roles.

    Risks and how to handle them

    • Writing slows shipping

      One post a month, rotating authors.

    Builds on

    Evidence

    • HiringFour open roles on Ashby, all engineering in San Francisco: Software Engineer, Infrastructure & Performance (posted September 29, 2026: the work "can bring product releases and customer commitments forward by weeks"; an "in-office homelab"; internal tools Hollywood, which "generates GitHub Actions from typed TypeScript definitions", and the Bessemer build system), Systems Engineer (posted June 29, 2026: the control plane and the Rust VMM, "the core technology our customers depend on"), a Systems Engineer Intern and an Infrastructure Engineer Intern. No marketing, sales or developer-relations roles are open.Source
    • GitHubGitHub organisation (84 public repos, 131 followers): the most-starred repos come from the earlier agent SDK: dedalus-agents-python (511 stars), dedalus-agents-typescript (284), wingman (180) and dedalus-mcp-python (138). The Machines tools are new and small: dedalus-cli (7), dedalus-typescript (5), dedalus-python (3) and dedalus-go (2), all updated November 6, 2026. The dedalus-agents-python README offers "convenient access to the Dedalus REST API" and links an MCP server, with no mention of Dedalus Machines.Source
    • Your websitellms.txt and llms-full.txt: "Dedalus Machines start in under 50ms" in the product section but "spins up in 50ms" in the company overview; full Linux with "systemd, package managers, and root access"; files "persist across sleep and wake"; a custom hypervisor built on Cloud Hypervisor; an FAQ that names E2B, Daytona and Fly.io without specifics; an agent skill installed with "npx skills add https://docs.dedaluslabs.ai --skill dedalus-machines"; an $11M seed and Y Combinator S25; no customers named.Source
  • 16

    The CLI and all three SDKs were all updated on November 6, 2026, but the site has no changelog and the blog's update posts were all re-dated to one day. Users and evaluators cannot see how fast the product moves, which is one of a small team's strongest arguments.

    Medium impactLow effort First result in 1 month Co Founder & CTO

    First action

    Collect the last eight weeks of release notes into a first page.

    How to measure it

    Reactivated accounts

    Now
    –
    Target
    10% of dormant accounts
    By
    1 quarter

    Steps

    1. 1

      Backfill

      Eight weeks of entries.

      Co-Founder & CTO half a day

    2. 2

      Automate

      Generate entries from tagged releases in CI.

      Infrastructure Engineer Intern 2 days

    3. 3

      Publish the page

      /changelog with RSS.

      Software Engineer, Infrastructure & Performance 1 day

    4. 4

      Link it

      Docs, footer, status page.

      Software Engineer, Infrastructure & Performance 1 hour

    5. 5

      Monthly email

      What shipped, to active and dormant users.

      Co-Founder & CEO monthly

    6. 6

      Measure

      Clicks and reactivations.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    $0
    Owner
    Co Founder & CTO

    Smallest test

    Send one "what shipped this month" email to all accounts.

    Build the monitor. A live page fed by the public registries and re-checked every day. Usually live within days.

    Book a working session

    Why it works

    Add /changelog with short dated entries generated from release notes, an RSS feed, and a monthly email to active users summarising what shipped. Link it from the docs, the footer and the status page.

    Visible shipping pace builds confidence for early adopters and brings dormant users back when a feature they wanted lands.

    Risks and how to handle them

    • Noise from small releases

      Group minor fixes into a weekly entry.

    Evidence

    • GitHubGitHub organisation (84 public repos, 131 followers): the most-starred repos come from the earlier agent SDK: dedalus-agents-python (511 stars), dedalus-agents-typescript (284), wingman (180) and dedalus-mcp-python (138). The Machines tools are new and small: dedalus-cli (7), dedalus-typescript (5), dedalus-python (3) and dedalus-go (2), all updated November 6, 2026. The dedalus-agents-python README offers "convenient access to the Dedalus REST API" and links an MCP server, with no mention of Dedalus Machines.Source
    • Your websiteSupport and trust: a Vanta trust center at trust.dedaluslabs.ai whose public page lists no frameworks; an incident.io status page (operational); a Discord community; support by email only, with no chat or meeting booking; no case studies, testimonials or changelog on the site.Source
    • Your websiteSitemap (24 URLs, November 9, 2026): 15 blog posts, six re-dated November 4, 2026 (seed round, Dedalus Auth launch, Machines update, "From Today to A2a", "What is an Agent", a YC interview); May 2026 posts from the earlier agent product ("Agent CRM Research", "Finance research", "One Click PDF Generator"); /store and /store/vault ("An interactive preview from the Dedalus Archive"); /private-beta. No comparison, use-case, customer-story or changelog pages. The blog description still reads "Technical writing, tutorials, and company updates about MCP, AI agents, and developer tooling." A /startup page ("Credits, support, deployment tooling, and marketplace distribution for startups building production AI agents on Dedalus") is live but missing from the sitemap and states no offer.Source
  • 17

    The careers page reaches crawlers with only a one-line description ("ship fast and solve hard problems") and no roles; the four engineering jobs exist only on Ashby. Search engines' job listings need the roles on the page with JobPosting data. The systems and infrastructure engineers Dedalus wants to hire are the same people who choose agent infrastructure at their own companies.

    Low impactLow effort First result in 2-4 weeks Co Founder & CTO

    First action

    Pull the Ashby job feed into /careers at build time.

    How to measure it

    Careers-page applications

    Now
    –
    Target
    2x
    By
    90 days

    Steps

    1. 1

      Pull roles

      Ashby job feed at build time.

      Infrastructure Engineer Intern 1 day

    2. 2

      Add structured data

      JobPosting for each role.

      Infrastructure Engineer Intern half a day

    3. 3

      Describe the work

      What each team builds, with links.

      Co-Founder & CTO half a day

    4. 4

      Add the product line

      Try a machine with the $20 credit.

      Co-Founder & CEO 10 minutes

    5. 5

      Validate

      Rich results test for each role.

      Infrastructure Engineer Intern 1 hour

    6. 6

      Measure

      Applications by source and sign-ups from /careers.

      Co-Founder & CTO monthly

    First result
    2-4 weeks
    Cost
    $0
    Owner
    Co Founder & CTO

    Smallest test

    List the four roles as plain links on /careers.

    Why it works

    Render the Ashby roles on /careers in the server HTML with JobPosting structured data, add what the team builds (hypervisor, Rust VMM, homelab) with links to the deep dives, and include a line inviting readers to try a machine with the $20 credit.

    A careers page that shows real engineering attracts candidates and doubles as a product page for a technical audience.

    Risks and how to handle them

    • Stale roles stay listed

      Rebuild the page whenever the Ashby feed changes.

    Builds on

    Evidence

    • Your websiteCareers page: the description says Dedalus is hiring engineers who "want to ship fast and solve hard problems", but the page a crawler receives holds no roles, mission or values; the roles appear only on Ashby.Source
    • HiringFour open roles on Ashby, all engineering in San Francisco: Software Engineer, Infrastructure & Performance (posted September 29, 2026: the work "can bring product releases and customer commitments forward by weeks"; an "in-office homelab"; internal tools Hollywood, which "generates GitHub Actions from typed TypeScript definitions", and the Bessemer build system), Systems Engineer (posted June 29, 2026: the control plane and the Rust VMM, "the core technology our customers depend on"), a Systems Engineer Intern and an Infrastructure Engineer Intern. No marketing, sales or developer-relations roles are open.Source

Big bets 3

Larger plays that can change the trajectory.

  • 18

    The site already has a /store section (with /store/vault, "An interactive preview from the Dedalus Archive") that today sells nothing a visitor can read. Machines are full Linux with root and persistent storage, so a configured machine is a reusable asset. E2B can fork a sandbox into up to 100 copies; Dedalus has no public way to start from someone else's setup.

    High impactHigh effort First result in 1 quarter Co Founder & CTO

    First action

    Decide what /store is for and write the template page spec.

    How to measure it

    Template-started machines

    Now
    –
    Target
    40% of new machines
    By
    2 quarters

    Steps

    1. 1

      Spec templates

      What a template captures and how it is versioned.

      Co-Founder & CTO 1 week

    2. 2

      Build publish and start

      CLI and API to publish and start from a template.

      Systems Engineer 6 weeks

    3. 3

      First-party set

      Four templates from the cookbook recipes.

      Systems Engineer Intern 3 weeks

    4. 4

      Template pages

      Server-rendered pages under /store.

      Software Engineer, Infrastructure & Performance 2 weeks

    5. 5

      Invite publishers

      Startup-program members first.

      Co-Founder & CEO 1 month

    6. 6

      Review and measure

      Security review of public templates; starts per template.

      Co-Founder & CTO ongoing

    First result
    1 quarter
    Cost
    A quarter of engineering time
    Owner
    Co Founder & CTO

    Smallest test

    Publish one first-party template with a one-line start command.

    Why it works

    Let users publish a machine image (OS, packages, agent code) as a template, and let others start a machine from it in one command. Launch with first-party templates (coding agent, browser agent, data science, CI) and invite startups from idea 6 to publish theirs. Each template page is a landing page with a one-line start command.

    Templates cut setup to seconds, create many search-friendly pages, and let users bring their own users to Dedalus.

    Risks and how to handle them

    • Unsafe public templates

      Review before listing and label community templates clearly.

    Builds on

    Evidence

    • Your websiteSitemap (24 URLs, November 9, 2026): 15 blog posts, six re-dated November 4, 2026 (seed round, Dedalus Auth launch, Machines update, "From Today to A2a", "What is an Agent", a YC interview); May 2026 posts from the earlier agent product ("Agent CRM Research", "Finance research", "One Click PDF Generator"); /store and /store/vault ("An interactive preview from the Dedalus Archive"); /private-beta. No comparison, use-case, customer-story or changelog pages. The blog description still reads "Technical writing, tutorials, and company updates about MCP, AI agents, and developer tooling." A /startup page ("Credits, support, deployment tooling, and marketplace distribution for startups building production AI agents on Dedalus") is live but missing from the sitemap and states no offer.Source
    • Competitor researchE2B homepage: "The AI Agent Cloud"; Firecracker microVMs with their own kernel; pause and resume from snapshots and fork into up to 100 copies; SOC 2 Type II; US, EU and APAC regions; Hobby with one-time $100 in usage credits and up to 1 hour of continuous runtime per sandbox; Pro at $150 a month plus usage with 24 hours of runtime.Source
    • DocsMachines are Linux VMs with their own kernel and root access that sleep "after five minutes of inactivity by default"; a CLI, an HTTP API and Python, TypeScript and Go SDKs; a four-step quickstart (install the CLI, set an API key, create a machine, run a command) with a default of 1 vCPU, 4,096 MiB and 10 GiB; "you only pay for the compute you use while the machine is running", with no unit rates. The docs index lists no cookbook, examples, pricing or limits pages, and a cookbook recipe still listed in search (agent-sandbox-per-user) returns 404.Source
  • 19

    Dedalus's earlier SDK drew over a thousand GitHub stars from agent builders, and the /startup page already promises "deployment tooling, and marketplace distribution". Products that host agents for their own users each need one isolated, persistent machine per user, which is exactly the pattern in Dedalus's own per-user recipe. One framework or product integration can bring thousands of machines.

    High impactHigh effort First result in 1-2 quarters Co Founder & CEO

    First action

    List the five frameworks and products whose users most need persistent machines.

    How to measure it

    Partner-created machines

    Now
    –
    Target
    25% of active machines
    By
    3 quarters

    Steps

    1. 1

      Shortlist

      Frameworks and products by fit and audience.

      Co-Founder & CEO 1 week

    2. 2

      Build the first integration

      A provider plug-in for one open-source framework.

      Systems Engineer 3 weeks

    3. 3

      Propose terms

      Volume pricing or revenue share.

      Co-Founder & CEO 1 month

    4. 4

      Co-launch

      Joint post, template and docs page.

      Co-Founder & CEO 2 weeks

    5. 5

      Support partners

      Shared channel with an engineer.

      Co-Founder & CTO ongoing

    6. 6

      Measure

      Machines and revenue per partner.

      Co-Founder & CEO monthly

    First result
    1-2 quarters
    Cost
    Engineering time and volume discounts
    Owner
    Co Founder & CEO

    Smallest test

    Ship one framework plug-in and announce it with the framework's maintainers.

    Why it works

    Build official integrations with two or three open-source agent frameworks and pitch three agent products that need a computer per user. Offer a revenue share or volume pricing, a co-written launch post and a template in the gallery. Measure machines created through each partner.

    Infrastructure that ships as a framework default grows with every developer who adopts the framework, at no extra sales cost.

    Risks and how to handle them

    • Partners build their own sandboxes

      Lead with persistence and free sleep, which are costly to build.

    Builds on

    Evidence

    • GitHubGitHub organisation (84 public repos, 131 followers): the most-starred repos come from the earlier agent SDK: dedalus-agents-python (511 stars), dedalus-agents-typescript (284), wingman (180) and dedalus-mcp-python (138). The Machines tools are new and small: dedalus-cli (7), dedalus-typescript (5), dedalus-python (3) and dedalus-go (2), all updated November 6, 2026. The dedalus-agents-python README offers "convenient access to the Dedalus REST API" and links an MCP server, with no mention of Dedalus Machines.Source
    • Your websiteSitemap (24 URLs, November 9, 2026): 15 blog posts, six re-dated November 4, 2026 (seed round, Dedalus Auth launch, Machines update, "From Today to A2a", "What is an Agent", a YC interview); May 2026 posts from the earlier agent product ("Agent CRM Research", "Finance research", "One Click PDF Generator"); /store and /store/vault ("An interactive preview from the Dedalus Archive"); /private-beta. No comparison, use-case, customer-story or changelog pages. The blog description still reads "Technical writing, tutorials, and company updates about MCP, AI agents, and developer tooling." A /startup page ("Credits, support, deployment tooling, and marketplace distribution for startups building production AI agents on Dedalus") is live but missing from the sitemap and states no offer.Source
    • DocsMachines are Linux VMs with their own kernel and root access that sleep "after five minutes of inactivity by default"; a CLI, an HTTP API and Python, TypeScript and Go SDKs; a four-step quickstart (install the CLI, set an API key, create a machine, run a command) with a default of 1 vCPU, 4,096 MiB and 10 GiB; "you only pay for the compute you use while the machine is running", with no unit rates. The docs index lists no cookbook, examples, pricing or limits pages, and a cookbook recipe still listed in search (agent-sandbox-per-user) returns 404.Source
  • 20

    Every provider quotes its own number (Dedalus under 50ms, Daytona sub-90ms creation, Blaxel about 25ms resume), each measuring something slightly different, and no neutral comparison exists. For the cold-start topic, search already returns Dedalus's own post next to rivals' guides. Dedalus has a homelab, a performance engineer role and the strongest speed claim, so it is well placed to define the measurement.

    High impactHigh effort First result in 1 quarter Co Founder & CTO

    First action

    Write the method: workloads, metrics, machine sizes and how runs are timed.

    How to measure it

    Referring domains to the benchmark

    Now
    –
    Target
    50
    By
    2 quarters

    Steps

    1. 1

      Write the method

      Workloads, metrics, sizes, timing.

      Co-Founder & CTO 1 week

    2. 2

      Build the harness

      Open source, one adapter per provider.

      Software Engineer, Infrastructure & Performance 3 weeks

    3. 3

      Run

      1,000 runs per provider, from two locations.

      Infrastructure Engineer Intern 1 week

    4. 4

      Invite review

      Share setup with each provider before publishing.

      Co-Founder & CEO 2 weeks

    5. 5

      Publish

      Results page, raw data and a write-up.

      Co-Founder & CTO 1 week

    6. 6

      Repeat

      Every quarter, with changes logged.

      Software Engineer, Infrastructure & Performance quarterly

    First result
    1 quarter
    Cost
    About a month of engineering time plus small compute costs
    Owner
    Co Founder & CTO

    Smallest test

    Benchmark Dedalus alone with the public harness and invite others to add adapters.

    Why it works

    Open-source a harness that measures start, wake, first-command latency, persistence and cost for a standard agent workload across providers, using each provider's public free tier. Publish results quarterly with full method and raw data, invite rivals to review their setup, and report honestly where Dedalus loses.

    The company that owns the benchmark frames the category; an honest one earns links, citations by AI assistants and credibility with engineers.

    Risks and how to handle them

    • Seen as biased

      Open code, raw data, rival review and published losses.

    • Provider terms restrict benchmarking

      Check each provider's terms and include only those that allow it.

    Builds on

    Evidence

    • Competitor researchDaytona homepage: "Run AI Code."; "Sub 90ms sandbox creation"; "$200 in free compute included"; published rates of $0.0504 per vCPU-hour and $0.0162 per GiB of memory per hour; nine named customer logos and six testimonials; HIPAA, SOC 2 and GDPR claims.Source
    • Competitor researchBlaxel homepage: "Infrastructure for autonomous agents"; about 25ms resume with memory state kept; up to $200 in credits with no card required; SOC 2, HIPAA and ISO 27001; 14 customer logos; the page says Blaxel has joined Baseten.Source
    • Search resultsWeb search, November 9, 2026: the homepage, /blog, /blog/sandbox-cold-start, /blog/dedalus-seed-round and /careers are listed under the title "DedalusDedalusDedalus Labs" instead of their page titles; for the cold-start topic, the other results are rivals' guides (Blaxel, Unikraft).Source
    • HiringFour open roles on Ashby, all engineering in San Francisco: Software Engineer, Infrastructure & Performance (posted September 29, 2026: the work "can bring product releases and customer commitments forward by weeks"; an "in-office homelab"; internal tools Hollywood, which "generates GitHub Actions from typed TypeScript definitions", and the Bessemer build system), Systems Engineer (posted June 29, 2026: the control plane and the Rust VMM, "the core technology our customers depend on"), a Systems Engineer Intern and an Infrastructure Engineer Intern. No marketing, sales or developer-relations roles are open.Source

Wildcards 2

Unexpected experiments, each with a cheap way to test it.

  • 21

    Dedalus already gives every new account a $20 credit without a card, has a Discord community and bills only active seconds, so a dollar of credit costs much less than a dollar to provide when much of the time is spent asleep. There is no referral offer anywhere on the site.

    Medium impactLow effort First result in 1 month Co Founder & CEO

    First action

    Set the reward amount and the qualifying event (first command run).

    How to measure it

    Referred sign-ups

    Now
    –
    Target
    15% of sign-ups
    By
    2 quarters

    Steps

    1. 1

      Set the reward

      Credit for both sides, with a cap.

      Co-Founder & CEO 1 hour

    2. 2

      Build links

      Referral codes on accounts.

      Software Engineer, Infrastructure & Performance 3 days

    3. 3

      Grant automatically

      Credit when the referred account runs a first command.

      Systems Engineer 2 days

    4. 4

      Surface it

      Dashboard, CLI message, Discord.

      Software Engineer, Infrastructure & Performance 1 day

    5. 5

      Guard abuse

      Limits per account and device checks.

      Co-Founder & CTO 2 days

    6. 6

      Measure

      Referred sign-ups and their paid conversion.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    Compute credits
    Owner
    Co Founder & CEO

    Smallest test

    Offer referral credit by hand in the Discord for a month.

    Why it works

    Give each user a referral link; when the referred developer runs their first command on a machine, both get extra credit. Show the link in the dashboard, CLI output after the first successful machine, and the Discord. Cap rewards per account.

    Developers recommend tools to teammates and friends, and paying in compute rewards only real usage.

    Risks and how to handle them

    • Fake accounts for credit

      Reward only after a real command runs, with caps.

    Builds on

    Evidence

    • Your websitePricing page: meta description "Dedalus Machines pricing: $20 sign-up credit, Pro at $20/month, and per-second billing for active compute. Idle compute is free."; Hobby $0 with a "$20 one-time sign-up credit. No credit card required."; Pro "$20/month with $20 in monthly compute credits"; Enterprise "Custom pricing, dedicated support, SLAs, custom limits" (llms-full.txt). Storage is "priced separately", and neither the site, llms-full.txt nor the docs publish a per-unit rate for vCPU, memory or storage.Source
    • Your websiteSupport and trust: a Vanta trust center at trust.dedaluslabs.ai whose public page lists no frameworks; an incident.io status page (operational); a Discord community; support by email only, with no chat or meeting booking; no case studies, testimonials or changelog on the site.Source
  • 22

    E2B's Hobby sandboxes run for up to 1 hour and Pro for 24 hours, while Dedalus machines keep their files across sleep, sleep for free and wake on demand. Agents increasingly run for days (research, per-user assistants, background coding agents), and those builders are hitting runtime caps. Dedalus says "sleep is free" but does not frame it as the answer to that pain.

    Medium impactLow effort First result in 1 month Co Founder & CEO

    First action

    Write the homepage line and the day-3 post outline.

    How to measure it

    Long-running use-case sign-ups

    Now
    –
    Target
    30% of sign-ups
    By
    2 quarters

    Steps

    1. 1

      Write the line

      Homepage and pricing.

      Co-Founder & CEO 1 hour

    2. 2

      Write the post

      Hour 2 and day 3 on capped versus persistent machines, with sources.

      Co-Founder & CEO 1 week

    3. 3

      Build the recipe

      An agent that resumes work across days.

      Systems Engineer Intern 1 week

    4. 4

      Add a sign-up question

      "How long do your agents run?"

      Software Engineer, Infrastructure & Performance 1 hour

    5. 5

      Distribute

      Forums and threads about sandbox time limits.

      Co-Founder & CEO 2 weeks

    6. 6

      Measure

      Answers and conversion by use case.

      Co-Founder & CEO monthly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Co Founder & CEO

    Smallest test

    A/B the homepage line for two weeks.

    Why it works

    Run a campaign around long-lived agents: a homepage line ("Your agent's computer is still there tomorrow"), a post comparing what happens at hour 2 and day 3 on capped sandboxes versus a persistent machine, a recipe for an agent that works across days, and targeted posts where developers discuss sandbox time limits.

    A sharp contrast on a felt pain beats a feature list, and it is a difference rivals cannot match by changing a price.

    Risks and how to handle them

    • Rivals lift their limits

      Lead with free sleep and kept files, not only the cap.

    Evidence

    • Competitor researchE2B homepage: "The AI Agent Cloud"; Firecracker microVMs with their own kernel; pause and resume from snapshots and fork into up to 100 copies; SOC 2 Type II; US, EU and APAC regions; Hobby with one-time $100 in usage credits and up to 1 hour of continuous runtime per sandbox; Pro at $150 a month plus usage with 24 hours of runtime.Source
    • Your websitePricing page: meta description "Dedalus Machines pricing: $20 sign-up credit, Pro at $20/month, and per-second billing for active compute. Idle compute is free."; Hobby $0 with a "$20 one-time sign-up credit. No credit card required."; Pro "$20/month with $20 in monthly compute credits"; Enterprise "Custom pricing, dedicated support, SLAs, custom limits" (llms-full.txt). Storage is "priced separately", and neither the site, llms-full.txt nor the docs publish a per-unit rate for vCPU, memory or storage.Source
    • Your websitellms.txt and llms-full.txt: "Dedalus Machines start in under 50ms" in the product section but "spins up in 50ms" in the company overview; full Linux with "systemd, package managers, and root access"; files "persist across sleep and wake"; a custom hypervisor built on Cloud Hypervisor; an FAQ that names E2B, Daytona and Fly.io without specifics; an agent skill installed with "npx skills add https://docs.dedaluslabs.ai --skill dedalus-machines"; an $11M seed and Y Combinator S25; no customers named.Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

Have it built

ProgrammableAgency

1 of the 22 ideas are things Programmable builds and runs for its customers: pages, outreach, prospect lists, booking flows and monitors. Pick one, or send the whole list.

Monitors1A live page fed by the public registries and re-checked every day.

A 30-minute session with the engineer who would build it, not a sales call. You leave with a plan whether or not you hire us.

Book a working session

Usually live within days.

Tools for the rest

ProgrammableDigital

Show it, don't just describe it

These ideas need something to show: a demo, a video, an animation, graphics or a deck. Digital turns screenshots or the site into all of them, in the company's own brand.

Make a demo (opens programmable.digital in a new tab)
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • ChillRep growth strategyA Dublin agency that scopes, builds and deploys one AI agent for one team in about four weeks, inside the client's own AWS, Azure or GCP…24 growth ideas
  • Gumloop growth strategyA platform where anyone at a company builds, shares and runs AI agents and workflows across hundreds of connectors and many AI models…22 growth ideas
  • ALX Africa growth strategyOnline and hub-based tech, AI, creative and entrepreneurship training for young Africans, with an All Access subscription for every…22 growth ideas
  • Icarus growth strategyBuilds APOLLO, a solar-powered autonomous aircraft that flies above 60,000 ft for persistent sensing, communications relay and data…22 growth ideas
  • Loyal growth strategyA clinical-stage veterinary medicine company developing drugs intended to extend the healthy lifespan of dogs (LOY-002 for senior dogs…21 growth ideas
  • The Orchard growth strategyA full-service music distribution and artist and label services company, part of Sony Music, offering distribution, marketing, sync…21 growth ideas
  • Quintagroup growth strategyA Python and JavaScript software development company from Lviv, Ukraine, known for e-procurement and auction systems (the OpenProcurement…23 growth ideas
  • Revise Robotics growth strategyBuilds AI-enabled robotic systems that refurbish used laptops end to end: a robot arm takes in each machine, tests it, wipes the drive…22 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on November 10, 2026, and cites them throughout. Dedalus Labs did not commission, review or endorse it, and Programmable is not affiliated with Dedalus Labs. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how Dedalus Labs runs its business.

Something wrong, or do you represent Dedalus Labs? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 307 ideas

    Make the site readable to crawlers, publish unit rates, settle one speed number, speed up the homepage, point the old SDK audience at Machines, clean up MCP-era pages and give /startup a real offer.

  2. Days 31 to 6012 ideas

    Comparison pages, first-machine activation, a recipe cookbook, the agent skill as a channel, trust and proof, an Enterprise path, engineering content, a changelog, careers pages, referral credits and the "computers that don't expire" message.

  3. Days 61 to 903 ideas

    Commit to the big bets: a template gallery, partner embeds and an open agent-compute benchmark.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value A full Linux computer per agent, with root and persistent files. The most cited problem: paying for idle or over-provisioned sandboxes.

  • Fast starts and free sleep, paying only for active seconds
  • Session time limits and lost state between runs
  • In their words: “sandbox”

Market

Competitive landscape

4 competitors compared: E2B, Daytona, Blaxel and Fly.io.

  • Gap to exploit: Plain, public unit pricing with a calculator for agent workloads
  • E2B: Hobby sandboxes run up to 1 hour, Pro up to 24 hours
  • Daytona: Sub-90ms creation, slower than Dedalus's stated start

Hiring signals

Dedalus is hiring only engineers, all in San Francisco: a Software Engineer for Infrastructure & Performance (developer tooling, build speed, a homelab), a Systems Engineer for the control plane and Rust VMM, and two…

  • Software Engineer, Infrastructure & Performance: Shipping speed, internal tooling and a homelab. Ideas 1, 7, 15, 20.
  • Systems Engineer: Control plane and Rust VMM that customers depend on. Ideas 3, 10, 18.
  • Infrastructure Engineer Intern: Development tools and workflows. Ideas 16, 17, 20.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks firstNew accounts get 100,000 free enrichment credits with code BETA100K (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammableMediaAn AI note taker, interpreter and webinar host for your calls (opens programmable.media in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence